ReasoningBank Intelligence

Pass

Audited by Gen Agent Trust Hub on Feb 28, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill implements an adaptive learning system that records experiences and learns patterns from external data, creating a surface for indirect prompt injection.
  • Ingestion points: Data enters the system via the rb.recordExperience and rb.learnPattern methods as documented in SKILL.md.
  • Boundary markers: The implementation lacks delimiters or explicit instructions to distinguish external context from system reasoning.
  • Capability inventory: The skill records data into a local database (./reasoning-bank.db) and utilizes this data to recommend future task strategies, which can influence agent behavior.
  • Sanitization: No evidence of input validation, filtering, or sanitization of ingested data is provided in the documentation or code examples.
Audit Metadata
Risk Level
SAFE
Analyzed
Feb 28, 2026, 10:51 AM